Lines Matching +full:oc +full:- +full:delay +full:- +full:us
1 // SPDX-License-Identifier: GPL-2.0-only
38 #include <linux/fault-inject.h>
61 /* Free Page Internal flags: for internal, non-pcp variants of free_pages(). */
70 * reporting it and marking it "reported" - it only skips notifying
79 * page shuffling (relevant code - e.g., memory onlining - is expected to
82 * Note: No code should rely on this flag for correctness - it's purely
89 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
102 /* UP spin_trylock always succeeds so disable IRQs to prevent re-entrancy. */
124 * Generic helper to lookup and a per-cpu variable with an embedded spinlock.
132 spin_lock(&_ret->member); \
141 if (!spin_trylock(&_ret->member)) { \
150 spin_unlock(&ptr->member); \
214 * other index - this ensures that it will be put on the correct CMA freelist.
218 return page->index; in get_pcppage_migratetype()
223 page->index = migratetype; in set_pcppage_migratetype()
235 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
236 * 1G machine -> (16M dma, 784M normal, 224M high)
241 * TBD: should special case ZONE_DMA32 machines here - in those we normally
289 int user_min_free_kbytes = -1;
314 * During boot we initialize deferred pages on-demand, as needed, but once
350 return page_zone(page)->pageblock_flags; in get_pageblock_bitmap()
357 pfn &= (PAGES_PER_SECTION-1); in pfn_to_bitidx()
359 pfn = pfn - pageblock_start_pfn(page_zone(page)->zone_start_pfn); in pfn_to_bitidx()
365 …* get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block o…
382 bitidx &= (BITS_PER_LONG-1); in get_pfnblock_flags_mask()
399 …* set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pag…
419 bitidx &= (BITS_PER_LONG-1); in set_pfnblock_flags_mask()
451 start_pfn = zone->zone_start_pfn; in page_outside_zone_boundaries()
452 sp = zone->spanned_pages; in page_outside_zone_boundaries()
457 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n", in page_outside_zone_boundaries()
458 pfn, zone_to_nid(zone), zone->name, in page_outside_zone_boundaries()
510 current->comm, page_to_pfn(page)); in bad_page()
569 * Higher-order pages are called "compound pages". They are structured thusly:
574 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
576 * The first tail page's ->compound_order holds the order of allocation.
577 * This usage means that zero-order pages may not be compound.
603 free_the_page(&folio->page, folio_order(folio)); in destroy_large_folio()
615 struct capture_control *capc = current->capture_control; in task_capc()
618 !(current->flags & PF_KTHREAD) && in task_capc()
619 !capc->page && in task_capc()
620 capc->cc->zone == zone ? capc : NULL; in task_capc()
627 if (!capc || order != capc->cc->order) in compaction_capture()
638 * and vice-versa but no more than normal fallback logic which can in compaction_capture()
639 * have trouble finding a high-order free page. in compaction_capture()
644 capc->page = page; in compaction_capture()
666 struct free_area *area = &zone->free_area[order]; in add_to_free_list()
668 list_add(&page->buddy_list, &area->free_list[migratetype]); in add_to_free_list()
669 area->nr_free++; in add_to_free_list()
676 struct free_area *area = &zone->free_area[order]; in add_to_free_list_tail()
678 list_add_tail(&page->buddy_list, &area->free_list[migratetype]); in add_to_free_list_tail()
679 area->nr_free++; in add_to_free_list_tail()
684 * of the list - so the moved pages won't immediately be considered for
690 struct free_area *area = &zone->free_area[order]; in move_to_free_list()
692 list_move_tail(&page->buddy_list, &area->free_list[migratetype]); in move_to_free_list()
702 list_del(&page->buddy_list); in del_page_from_free_list()
705 zone->free_area[order].nr_free--; in del_page_from_free_list()
711 return list_first_entry_or_null(&area->free_list[migratetype], in get_page_from_free_area()
717 * of the next-highest order is free. If it is, it's possible
730 if (order >= MAX_PAGE_ORDER - 1) in buddy_merge_likely()
734 higher_page = page + (higher_page_pfn - pfn); in buddy_merge_likely()
743 * The concept of a buddy system is to maintain direct-mapped table
761 * -- nyc
776 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page); in __free_one_page()
778 VM_BUG_ON(migratetype == -1); in __free_one_page()
782 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page); in __free_one_page()
787 __mod_zone_freepage_state(zone, -(1 << order), in __free_one_page()
820 page = page + (combined_pfn - pfn); in __free_one_page()
846 * split_free_page() -- split a free page at split_pfn_offset
851 * Return -ENOENT if the free page is changed, otherwise 0
872 spin_lock_irqsave(&zone->lock, flags); in split_free_page()
875 ret = -ENOENT; in split_free_page()
881 __mod_zone_freepage_state(zone, -(1UL << order), mt); in split_free_page()
894 split_pfn_offset -= (1UL << free_page_order); in split_free_page()
897 split_pfn_offset = (1UL << order) - (pfn - free_page_pfn); in split_free_page()
900 spin_unlock_irqrestore(&zone->lock, flags); in split_free_page()
911 if (unlikely(atomic_read(&page->_mapcount) != -1)) in page_expected_state()
914 if (unlikely((unsigned long)page->mapping | in page_expected_state()
917 page->memcg_data | in page_expected_state()
920 ((page->pp_magic & ~0x3UL) == PP_SIGNATURE) | in page_expected_state()
922 (page->flags & check_flags))) in page_expected_state()
932 if (unlikely(atomic_read(&page->_mapcount) != -1)) in page_bad_reason()
934 if (unlikely(page->mapping != NULL)) in page_bad_reason()
935 bad_reason = "non-NULL mapping"; in page_bad_reason()
938 if (unlikely(page->flags & flags)) { in page_bad_reason()
945 if (unlikely(page->memcg_data)) in page_bad_reason()
949 if (unlikely((page->pp_magic & ~0x3UL) == PP_SIGNATURE)) in page_bad_reason()
982 * We rely page->lru.next never has bit 0 set, unless the page in free_tail_page_prepare()
983 * is PageTail(). Let's make sure that's true even for poisoned ->lru. in free_tail_page_prepare()
991 switch (page - head_page) { in free_tail_page_prepare()
993 /* the first tail page: these may be in place of ->mapping */ in free_tail_page_prepare()
998 if (unlikely(atomic_read(&folio->_nr_pages_mapped))) { in free_tail_page_prepare()
1002 if (unlikely(atomic_read(&folio->_pincount))) { in free_tail_page_prepare()
1009 * the second tail page: ->mapping is in free_tail_page_prepare()
1010 * deferred_list.next -- ignore value. in free_tail_page_prepare()
1014 if (page->mapping != TAIL_MAPPING) { in free_tail_page_prepare()
1030 page->mapping = NULL; in free_tail_page_prepare()
1039 * Tag-based KASAN modes skip pages freed via deferred memory initialization
1041 * 2. For tag-based KASAN modes: the page has a match-all KASAN tag, indicating
1044 * Pages will have match-all tags in the following circumstances:
1061 * on-demand allocation and then freed again before the deferred pages
1110 * avoid checking PageCompound for order-0 pages. in free_pages_prepare()
1126 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; in free_pages_prepare()
1130 page->mapping = NULL; in free_pages_prepare()
1139 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP; in free_pages_prepare()
1157 * With hardware tag-based KASAN, memory tags must be set before the in free_pages_prepare()
1200 count = min(pcp->count, count); in free_pcppages_bulk()
1203 pindex = pindex - 1; in free_pcppages_bulk()
1205 spin_lock_irqsave(&zone->lock, flags); in free_pcppages_bulk()
1212 /* Remove pages from lists in a round-robin fashion. */ in free_pcppages_bulk()
1214 if (++pindex > NR_PCP_LISTS - 1) in free_pcppages_bulk()
1216 list = &pcp->lists[pindex]; in free_pcppages_bulk()
1228 list_del(&page->pcp_list); in free_pcppages_bulk()
1229 count -= nr_pages; in free_pcppages_bulk()
1230 pcp->count -= nr_pages; in free_pcppages_bulk()
1243 spin_unlock_irqrestore(&zone->lock, flags); in free_pcppages_bulk()
1253 spin_lock_irqsave(&zone->lock, flags); in free_one_page()
1259 spin_unlock_irqrestore(&zone->lock, flags); in free_one_page()
1296 for (loop = 0; loop < (nr_pages - 1); loop++, p++) { in __free_pages_core()
1304 atomic_long_add(nr_pages, &page_zone(page)->managed_pages); in __free_pages_core()
1335 * Note: the function may return non-NULL struct page even for a page block
1338 * will fall into 2 sub-sections, and the end pfn of the pageblock may be hole
1351 end_pfn--; in __pageblock_pfn_to_page()
1384 * -- nyc
1392 high--; in expand()
1412 if (unlikely(page->flags & __PG_HWPOISON)) { in check_new_page_bad()
1456 /* Skip, if hardware tag-based KASAN is not enabled. */ in should_skip_kasan_unpoison()
1461 * With hardware tag-based KASAN enabled, skip if this has been in should_skip_kasan_unpoison()
1469 /* Don't skip, if hardware tag-based KASAN is not enabled. */ in should_skip_init()
1473 /* For hardware tag-based KASAN, skip if requested. */ in should_skip_init()
1571 area = &(zone->free_area[current_order]); in __rmqueue_smallest()
1594 static int fallbacks[MIGRATE_TYPES][MIGRATE_PCPTYPES - 1] = {
1663 end_pfn = pageblock_end_pfn(pfn) - 1; in move_freepages_block()
1678 int nr_pageblocks = 1 << (start_order - pageblock_order); in change_pageblock_range()
1680 while (nr_pageblocks--) { in change_pageblock_range()
1734 max_boost = mult_frac(zone->_watermark[WMARK_HIGH], in boost_watermark()
1750 zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages, in boost_watermark()
1759 * pageblock to our migratetype and determine how many already-allocated pages
1792 set_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); in steal_suitable_fallback()
1814 * to MOVABLE pageblock, consider all non-movable pages as in steal_suitable_fallback()
1817 * exact migratetype of non-movable pages. in steal_suitable_fallback()
1821 - (free_pages + movable_pages); in steal_suitable_fallback()
1829 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) || in steal_suitable_fallback()
1851 if (area->nr_free == 0) in find_suitable_fallback()
1852 return -1; in find_suitable_fallback()
1855 for (i = 0; i < MIGRATE_PCPTYPES - 1 ; i++) { in find_suitable_fallback()
1870 return -1; in find_suitable_fallback()
1874 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1886 * Check is race-prone but harmless. in reserve_highatomic_pageblock()
1891 if (zone->nr_reserved_highatomic >= max_managed) in reserve_highatomic_pageblock()
1894 spin_lock_irqsave(&zone->lock, flags); in reserve_highatomic_pageblock()
1897 if (zone->nr_reserved_highatomic >= max_managed) in reserve_highatomic_pageblock()
1904 zone->nr_reserved_highatomic += pageblock_nr_pages; in reserve_highatomic_pageblock()
1910 spin_unlock_irqrestore(&zone->lock, flags); in reserve_highatomic_pageblock()
1915 * potentially hurts the reliability of high-order allocations when under
1925 struct zonelist *zonelist = ac->zonelist; in unreserve_highatomic_pageblock()
1933 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->highest_zoneidx, in unreserve_highatomic_pageblock()
1934 ac->nodemask) { in unreserve_highatomic_pageblock()
1939 if (!force && zone->nr_reserved_highatomic <= in unreserve_highatomic_pageblock()
1943 spin_lock_irqsave(&zone->lock, flags); in unreserve_highatomic_pageblock()
1945 struct free_area *area = &(zone->free_area[order]); in unreserve_highatomic_pageblock()
1955 * from highatomic to ac->migratetype. So we should in unreserve_highatomic_pageblock()
1961 * locking could inadvertently allow a per-cpu in unreserve_highatomic_pageblock()
1966 zone->nr_reserved_highatomic -= min( in unreserve_highatomic_pageblock()
1968 zone->nr_reserved_highatomic); in unreserve_highatomic_pageblock()
1972 * Convert to ac->migratetype and avoid the normal in unreserve_highatomic_pageblock()
1980 set_pageblock_migratetype(page, ac->migratetype); in unreserve_highatomic_pageblock()
1981 ret = move_freepages_block(zone, page, ac->migratetype, in unreserve_highatomic_pageblock()
1984 spin_unlock_irqrestore(&zone->lock, flags); in unreserve_highatomic_pageblock()
1988 spin_unlock_irqrestore(&zone->lock, flags); in unreserve_highatomic_pageblock()
2029 --current_order) { in __rmqueue_fallback()
2030 area = &(zone->free_area[current_order]); in __rmqueue_fallback()
2033 if (fallback_mt == -1) in __rmqueue_fallback()
2055 area = &(zone->free_area[current_order]); in __rmqueue_fallback()
2058 if (fallback_mt != -1) in __rmqueue_fallback()
2063 * This should not happen - we already found a suitable fallback in __rmqueue_fallback()
2083 * Call me with the zone->lock already held.
2130 spin_lock_irqsave(&zone->lock, flags); in rmqueue_bulk()
2147 list_add_tail(&page->pcp_list, list); in rmqueue_bulk()
2150 -(1 << order)); in rmqueue_bulk()
2153 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order)); in rmqueue_bulk()
2154 spin_unlock_irqrestore(&zone->lock, flags); in rmqueue_bulk()
2168 high_min = READ_ONCE(pcp->high_min); in decay_pcp_high()
2169 batch = READ_ONCE(pcp->batch); in decay_pcp_high()
2171 * Decrease pcp->high periodically to try to free possible in decay_pcp_high()
2173 * control latency. This caps pcp->high decrement too. in decay_pcp_high()
2175 if (pcp->high > high_min) { in decay_pcp_high()
2176 pcp->high = max3(pcp->count - (batch << CONFIG_PCP_BATCH_SCALE_MAX), in decay_pcp_high()
2177 pcp->high - (pcp->high >> 3), high_min); in decay_pcp_high()
2178 if (pcp->high > high_min) in decay_pcp_high()
2182 to_drain = pcp->count - pcp->high; in decay_pcp_high()
2184 spin_lock(&pcp->lock); in decay_pcp_high()
2186 spin_unlock(&pcp->lock); in decay_pcp_high()
2203 batch = READ_ONCE(pcp->batch); in drain_zone_pages()
2204 to_drain = min(pcp->count, batch); in drain_zone_pages()
2206 spin_lock(&pcp->lock); in drain_zone_pages()
2208 spin_unlock(&pcp->lock); in drain_zone_pages()
2220 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); in drain_pages_zone()
2221 if (pcp->count) { in drain_pages_zone()
2222 spin_lock(&pcp->lock); in drain_pages_zone()
2223 free_pcppages_bulk(zone, pcp->count, pcp, 0); in drain_pages_zone()
2224 spin_unlock(&pcp->lock); in drain_pages_zone()
2241 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2258 * not empty. The check for non-emptiness can however race with a free to
2259 * pcplist that has not yet increased the pcp->count from 0 to 1. Callers
2302 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); in __drain_all_pages()
2303 if (pcp->count) in __drain_all_pages()
2307 pcp = per_cpu_ptr(z->per_cpu_pageset, cpu); in __drain_all_pages()
2308 if (pcp->count) { in __drain_all_pages()
2332 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2334 * When zone parameter is non-NULL, spill just the single zone's pages.
2358 /* Free as much as possible if batch freeing high-order pages. */ in nr_pcp_free()
2360 return min(pcp->count, batch << CONFIG_PCP_BATCH_SCALE_MAX); in nr_pcp_free()
2366 /* Leave at least pcp->batch pages on the list */ in nr_pcp_free()
2368 max_nr_free = high - batch; in nr_pcp_free()
2374 batch = clamp_t(int, pcp->free_count, min_nr_free, max_nr_free); in nr_pcp_free()
2384 high_min = READ_ONCE(pcp->high_min); in nr_pcp_high()
2385 high_max = READ_ONCE(pcp->high_max); in nr_pcp_high()
2386 high = pcp->high = clamp(pcp->high, high_min, high_max); in nr_pcp_high()
2392 pcp->high = max(high - (batch << CONFIG_PCP_BATCH_SCALE_MAX), in nr_pcp_high()
2401 if (test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) { in nr_pcp_high()
2402 int free_count = max_t(int, pcp->free_count, batch); in nr_pcp_high()
2404 pcp->high = max(high - free_count, high_min); in nr_pcp_high()
2405 return min(batch << 2, pcp->high); in nr_pcp_high()
2411 if (test_bit(ZONE_BELOW_HIGH, &zone->flags)) { in nr_pcp_high()
2412 int free_count = max_t(int, pcp->free_count, batch); in nr_pcp_high()
2414 pcp->high = max(high - free_count, high_min); in nr_pcp_high()
2415 high = max(pcp->count, high_min); in nr_pcp_high()
2416 } else if (pcp->count >= high) { in nr_pcp_high()
2417 int need_high = pcp->free_count + batch; in nr_pcp_high()
2419 /* pcp->high should be large enough to hold batch freed pages */ in nr_pcp_high()
2420 if (pcp->high < need_high) in nr_pcp_high()
2421 pcp->high = clamp(need_high, high_min, high_max); in nr_pcp_high()
2440 pcp->alloc_factor >>= 1; in free_unref_page_commit()
2443 list_add(&page->pcp_list, &pcp->lists[pindex]); in free_unref_page_commit()
2444 pcp->count += 1 << order; in free_unref_page_commit()
2446 batch = READ_ONCE(pcp->batch); in free_unref_page_commit()
2448 * As high-order pages other than THP's stored on PCP can contribute in free_unref_page_commit()
2454 free_high = (pcp->free_count >= batch && in free_unref_page_commit()
2455 (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) && in free_unref_page_commit()
2456 (!(pcp->flags & PCPF_FREE_HIGH_BATCH) || in free_unref_page_commit()
2457 pcp->count >= READ_ONCE(batch))); in free_unref_page_commit()
2458 pcp->flags |= PCPF_PREV_FREE_HIGH_ORDER; in free_unref_page_commit()
2459 } else if (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) { in free_unref_page_commit()
2460 pcp->flags &= ~PCPF_PREV_FREE_HIGH_ORDER; in free_unref_page_commit()
2462 if (pcp->free_count < (batch << CONFIG_PCP_BATCH_SCALE_MAX)) in free_unref_page_commit()
2463 pcp->free_count += (1 << order); in free_unref_page_commit()
2465 if (pcp->count >= high) { in free_unref_page_commit()
2468 if (test_bit(ZONE_BELOW_HIGH, &zone->flags) && in free_unref_page_commit()
2471 clear_bit(ZONE_BELOW_HIGH, &zone->flags); in free_unref_page_commit()
2507 pcp = pcp_spin_trylock(zone->per_cpu_pageset); in free_unref_page()
2518 * Free a list of 0-order pages
2533 list_del(&page->lru); in free_unref_page_list()
2543 list_del(&page->lru); in free_unref_page_list()
2552 list_del(&page->lru); in free_unref_page_list()
2573 pcp = pcp_spin_trylock(zone->per_cpu_pageset); in free_unref_page_list()
2585 * Non-isolated types over MIGRATE_PCPTYPES get added in free_unref_page_list()
2603 * split_page takes a non-compound higher-order page, and splits it into
2604 * n (1<<order) sub-pages: page[0..n]
2605 * Each sub-page must be freed individually.
2633 * emulate a high-order watermark check with a raised order-0 in __isolate_free_page()
2634 * watermark, because we already know our high-order page in __isolate_free_page()
2637 watermark = zone->_watermark[WMARK_MIN] + (1UL << order); in __isolate_free_page()
2641 __mod_zone_freepage_state(zone, -(1UL << order), mt); in __isolate_free_page()
2650 if (order >= pageblock_order - 1) { in __isolate_free_page()
2651 struct page *endpage = page + (1 << order) - 1; in __isolate_free_page()
2668 * __putback_isolated_page - Return a now-isolated page back where we got it
2681 lockdep_assert_held(&zone->lock); in __putback_isolated_page()
2724 spin_lock_irqsave(&zone->lock, flags); in rmqueue_buddy()
2733 * failing a high-order atomic allocation in the in rmqueue_buddy()
2740 spin_unlock_irqrestore(&zone->lock, flags); in rmqueue_buddy()
2744 __mod_zone_freepage_state(zone, -(1 << order), in rmqueue_buddy()
2746 spin_unlock_irqrestore(&zone->lock, flags); in rmqueue_buddy()
2760 base_batch = READ_ONCE(pcp->batch); in nr_pcp_alloc()
2761 high_min = READ_ONCE(pcp->high_min); in nr_pcp_alloc()
2762 high_max = READ_ONCE(pcp->high_max); in nr_pcp_alloc()
2763 high = pcp->high = clamp(pcp->high, high_min, high_max); in nr_pcp_alloc()
2772 batch = (base_batch << pcp->alloc_factor); in nr_pcp_alloc()
2775 * If we had larger pcp->high, we could avoid to allocate from in nr_pcp_alloc()
2778 if (high_min != high_max && !test_bit(ZONE_BELOW_HIGH, &zone->flags)) in nr_pcp_alloc()
2779 high = pcp->high = min(high + batch, high_max); in nr_pcp_alloc()
2782 max_nr_alloc = max(high - pcp->count - base_batch, base_batch); in nr_pcp_alloc()
2785 * subsequent allocation of order-0 pages without any freeing. in nr_pcp_alloc()
2788 pcp->alloc_factor < CONFIG_PCP_BATCH_SCALE_MAX) in nr_pcp_alloc()
2789 pcp->alloc_factor++; in nr_pcp_alloc()
2805 /* Remove page from the per-cpu list, caller must protect the list */
2824 pcp->count += alloced << order; in __rmqueue_pcplist()
2830 list_del(&page->pcp_list); in __rmqueue_pcplist()
2831 pcp->count -= 1 << order; in __rmqueue_pcplist()
2837 /* Lock and remove page from the per-cpu list */
2849 pcp = pcp_spin_trylock(zone->per_cpu_pageset); in rmqueue_pcplist()
2860 pcp->free_count >>= 1; in rmqueue_pcplist()
2861 list = &pcp->lists[order_to_pindex(migratetype, order)]; in rmqueue_pcplist()
2874 * Use pcplists for THP or "cheap" high-order allocations.
2894 * allocate greater than order-1 page units with __GFP_NOFAIL. in rmqueue()
2911 unlikely(test_bit(ZONE_BOOSTED_WATERMARK, &zone->flags))) { in rmqueue()
2912 clear_bit(ZONE_BOOSTED_WATERMARK, &zone->flags); in rmqueue()
2929 long unusable_free = (1 << order) - 1; in __zone_watermark_unusable_free()
2933 * watermark then subtract the high-atomic reserves. This will in __zone_watermark_unusable_free()
2934 * over-estimate the size of the atomic reserve but it avoids a search. in __zone_watermark_unusable_free()
2937 unusable_free += z->nr_reserved_highatomic; in __zone_watermark_unusable_free()
2952 * Return true if free base pages are above 'mark'. For high-order checks it
2953 * will return true of the order-0 watermark is reached and there is at least
2964 /* free_pages may go negative - that's OK */ in __zone_watermark_ok()
2965 free_pages -= __zone_watermark_unusable_free(z, order, alloc_flags); in __zone_watermark_ok()
2973 min -= min / 2; in __zone_watermark_ok()
2976 * Non-blocking allocations (e.g. GFP_ATOMIC) can in __zone_watermark_ok()
2978 * non-blocking allocations requests such as GFP_NOWAIT in __zone_watermark_ok()
2983 min -= min / 4; in __zone_watermark_ok()
2990 * makes during the free path will be small and short-lived. in __zone_watermark_ok()
2993 min -= min / 2; in __zone_watermark_ok()
2997 * Check watermarks for an order-0 allocation request. If these in __zone_watermark_ok()
2998 * are not met, then a high-order request also cannot go ahead in __zone_watermark_ok()
3001 if (free_pages <= min + z->lowmem_reserve[highest_zoneidx]) in __zone_watermark_ok()
3004 /* If this is an order-0 request then the watermark is fine */ in __zone_watermark_ok()
3008 /* For a high-order request, check at least one suitable page is free */ in __zone_watermark_ok()
3010 struct free_area *area = &z->free_area[o]; in __zone_watermark_ok()
3013 if (!area->nr_free) in __zone_watermark_ok()
3051 * Fast check for order-0 only. If this fails then the reserves in zone_watermark_fast()
3061 /* reserved may over estimate high-atomic reserves. */ in zone_watermark_fast()
3062 usable_free -= min(usable_free, reserved); in zone_watermark_fast()
3063 if (usable_free > mark + z->lowmem_reserve[highest_zoneidx]) in zone_watermark_fast()
3072 * Ignore watermark boosting for __GFP_HIGH order-0 allocations in zone_watermark_fast()
3077 if (unlikely(!order && (alloc_flags & ALLOC_MIN_RESERVE) && z->watermark_boost in zone_watermark_fast()
3079 mark = z->_watermark[WMARK_MIN]; in zone_watermark_fast()
3092 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark) in zone_watermark_ok_safe()
3142 * the pointer is within zone->zone_pgdat->node_zones[]. Also assume in alloc_flags_nofragment()
3145 BUILD_BUG_ON(ZONE_NORMAL - ZONE_DMA32 != 1); in alloc_flags_nofragment()
3146 if (nr_online_nodes > 1 && !populated_zone(--zone)) in alloc_flags_nofragment()
3185 z = ac->preferred_zoneref; in get_page_from_freelist()
3186 for_next_zone_zonelist_nodemask(zone, z, ac->highest_zoneidx, in get_page_from_freelist()
3187 ac->nodemask) { in get_page_from_freelist()
3206 * exceed the per-node dirty limit in the slowpath in get_page_from_freelist()
3212 * dirty-throttling and the flusher threads. in get_page_from_freelist()
3214 if (ac->spread_dirty_pages) { in get_page_from_freelist()
3215 if (last_pgdat != zone->zone_pgdat) { in get_page_from_freelist()
3216 last_pgdat = zone->zone_pgdat; in get_page_from_freelist()
3217 last_pgdat_dirty_ok = node_dirty_ok(zone->zone_pgdat); in get_page_from_freelist()
3225 zone != ac->preferred_zoneref->zone) { in get_page_from_freelist()
3233 local_nid = zone_to_nid(ac->preferred_zoneref->zone); in get_page_from_freelist()
3242 * watermark. If so, we will decrease pcp->high and free in get_page_from_freelist()
3247 if (test_bit(ZONE_BELOW_HIGH, &zone->flags)) in get_page_from_freelist()
3252 ac->highest_zoneidx, alloc_flags, in get_page_from_freelist()
3256 set_bit(ZONE_BELOW_HIGH, &zone->flags); in get_page_from_freelist()
3261 ac->highest_zoneidx, alloc_flags, in get_page_from_freelist()
3286 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone)) in get_page_from_freelist()
3289 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order); in get_page_from_freelist()
3300 ac->highest_zoneidx, alloc_flags)) in get_page_from_freelist()
3308 page = rmqueue(ac->preferred_zoneref->zone, zone, order, in get_page_from_freelist()
3309 gfp_mask, alloc_flags, ac->migratetype); in get_page_from_freelist()
3314 * If this is a high-order atomic allocation then check in get_page_from_freelist()
3360 (current->flags & (PF_MEMALLOC | PF_EXITING))) in warn_alloc_show_mem()
3383 current->comm, &vaf, gfp_mask, &gfp_mask, in warn_alloc()
3417 struct oom_control oc = { in __alloc_pages_may_oom() local
3418 .zonelist = ac->zonelist, in __alloc_pages_may_oom()
3419 .nodemask = ac->nodemask, in __alloc_pages_may_oom()
3430 * making progress for us. in __alloc_pages_may_oom()
3452 if (current->flags & PF_DUMPCORE) in __alloc_pages_may_oom()
3468 if (ac->highest_zoneidx < ZONE_NORMAL) in __alloc_pages_may_oom()
3483 if (out_of_memory(&oc) || in __alloc_pages_may_oom()
3488 * Help non-failing allocations by giving them access to memory in __alloc_pages_may_oom()
3507 /* Try memory compaction for high-order allocations before reclaim */
3550 zone->compact_blockskip_flush = false; in __alloc_pages_direct_compact()
3586 * Compaction was skipped due to a lack of free order-0 in should_compact_retry()
3624 (*compact_priority)--; in should_compact_retry()
3657 * Let's give them a good hope and keep retrying while the order-0 in should_compact_retry()
3660 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, in should_compact_retry()
3661 ac->highest_zoneidx, ac->nodemask) { in should_compact_retry()
3663 ac->highest_zoneidx, alloc_flags)) in should_compact_retry()
3681 if (current->flags & PF_MEMALLOC) in __need_reclaim()
3768 progress = try_to_free_pages(ac->zonelist, order, gfp_mask, in __perform_reclaim()
3769 ac->nodemask); in __perform_reclaim()
3799 * pages are pinned on the per-cpu lists or in high alloc reserves. in __alloc_pages_direct_reclaim()
3820 enum zone_type highest_zoneidx = ac->highest_zoneidx; in wake_all_kswapds()
3822 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, highest_zoneidx, in wake_all_kswapds()
3823 ac->nodemask) { in wake_all_kswapds()
3826 if (last_pgdat != zone->zone_pgdat) { in wake_all_kswapds()
3828 last_pgdat = zone->zone_pgdat; in wake_all_kswapds()
3868 * Ignore cpuset mems for non-blocking __GFP_HIGH (probably in gfp_to_alloc_flags()
3907 if (in_serving_softirq() && (current->flags & PF_MEMALLOC)) in __gfp_pfmemalloc_flags()
3910 if (current->flags & PF_MEMALLOC) in __gfp_pfmemalloc_flags()
3963 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, in should_reclaim_retry()
3964 ac->highest_zoneidx, ac->nodemask) { in should_reclaim_retry()
3978 ac->highest_zoneidx, alloc_flags, available); in should_reclaim_retry()
3994 if (current->flags & PF_WQ_WORKER) in should_reclaim_retry()
4015 * This assumes that for all allocations, ac->nodemask can come only in check_retry_cpuset()
4020 if (cpusets_enabled() && ac->nodemask && in check_retry_cpuset()
4021 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) { in check_retry_cpuset()
4022 ac->nodemask = NULL; in check_retry_cpuset()
4074 * there was a cpuset modification and we are retrying - otherwise we in __alloc_pages_slowpath()
4075 * could end up iterating over non-eligible zones endlessly. in __alloc_pages_slowpath()
4077 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, in __alloc_pages_slowpath()
4078 ac->highest_zoneidx, ac->nodemask); in __alloc_pages_slowpath()
4079 if (!ac->preferred_zoneref->zone) in __alloc_pages_slowpath()
4084 * any suitable zone to satisfy the request - e.g. non-movable in __alloc_pages_slowpath()
4088 struct zoneref *z = first_zones_zonelist(ac->zonelist, in __alloc_pages_slowpath()
4089 ac->highest_zoneidx, in __alloc_pages_slowpath()
4091 if (!z->zone) in __alloc_pages_slowpath()
4108 * that we have enough base pages and don't need to reclaim. For non- in __alloc_pages_slowpath()
4109 * movable high-order allocations, do that as well, as compaction will in __alloc_pages_slowpath()
4117 (order > 0 && ac->migratetype != MIGRATE_MOVABLE)) in __alloc_pages_slowpath()
4139 * - potentially very expensive because zones are far in __alloc_pages_slowpath()
4142 * - not guaranteed to help because isolate_freepages() in __alloc_pages_slowpath()
4145 * - unlikely to make entire pageblocks free on its in __alloc_pages_slowpath()
4177 ac->nodemask = NULL; in __alloc_pages_slowpath()
4178 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, in __alloc_pages_slowpath()
4179 ac->highest_zoneidx, ac->nodemask); in __alloc_pages_slowpath()
4192 if (current->flags & PF_MEMALLOC) in __alloc_pages_slowpath()
4224 * It doesn't make any sense to retry for the compaction if the order-0 in __alloc_pages_slowpath()
4244 /* Reclaim has failed us, start killing things */ in __alloc_pages_slowpath()
4285 * for somebody to do a work for us in __alloc_pages_slowpath()
4287 WARN_ON_ONCE_GFP(current->flags & PF_MEMALLOC, gfp_mask); in __alloc_pages_slowpath()
4298 * Help non-failing allocations by giving some access to memory in __alloc_pages_slowpath()
4299 * reserves normally used for high priority non-blocking in __alloc_pages_slowpath()
4312 warn_alloc(gfp_mask, ac->nodemask, in __alloc_pages_slowpath()
4323 ac->highest_zoneidx = gfp_zone(gfp_mask); in prepare_alloc_pages()
4324 ac->zonelist = node_zonelist(preferred_nid, gfp_mask); in prepare_alloc_pages()
4325 ac->nodemask = nodemask; in prepare_alloc_pages()
4326 ac->migratetype = gfp_migratetype(gfp_mask); in prepare_alloc_pages()
4334 if (in_task() && !ac->nodemask) in prepare_alloc_pages()
4335 ac->nodemask = &cpuset_current_mems_allowed; in prepare_alloc_pages()
4348 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE); in prepare_alloc_pages()
4355 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist, in prepare_alloc_pages()
4356 ac->highest_zoneidx, ac->nodemask); in prepare_alloc_pages()
4362 * __alloc_pages_bulk - Allocate a number of order-0 pages to a list or array
4409 if (unlikely(page_array && nr_pages - nr_populated == 0)) in __alloc_pages_bulk()
4417 if (nr_pages - nr_populated == 1) in __alloc_pages_bulk()
4448 if (nr_online_nodes > 1 && zone != ac.preferred_zoneref->zone && in __alloc_pages_bulk()
4449 zone_to_nid(zone) != zone_to_nid(ac.preferred_zoneref->zone)) { in __alloc_pages_bulk()
4470 pcp = pcp_spin_trylock(zone->per_cpu_pageset); in __alloc_pages_bulk()
4475 pcp_list = &pcp->lists[order_to_pindex(ac.migratetype, 0)]; in __alloc_pages_bulk()
4498 list_add(&page->lru, page_list); in __alloc_pages_bulk()
4508 zone_statistics(ac.preferred_zoneref->zone, zone, nr_account); in __alloc_pages_bulk()
4520 list_add(&page->lru, page_list); in __alloc_pages_bulk()
4566 alloc_flags |= alloc_flags_nofragment(ac.preferred_zoneref->zone, gfp); in __alloc_pages()
4578 * &cpuset_current_mems_allowed to optimize the fast-path attempt. in __alloc_pages()
4630 * __free_pages - Free pages allocated with alloc_pages().
4634 * This function can free multi-page allocations that are not compound
4640 * by put_page() which only frees the first page of a non-compound
4657 while (order-- > 0) in __free_pages()
4674 * An arbitrary-length arbitrary-offset area of memory which resides
4681 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4694 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE; in __page_frag_cache_refill()
4699 nc->va = page ? page_address(page) : NULL; in __page_frag_cache_refill()
4721 if (unlikely(!nc->va)) { in page_frag_alloc_align()
4729 size = nc->size; in page_frag_alloc_align()
4737 nc->pfmemalloc = page_is_pfmemalloc(page); in page_frag_alloc_align()
4738 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; in page_frag_alloc_align()
4739 nc->offset = size; in page_frag_alloc_align()
4742 offset = nc->offset - fragsz; in page_frag_alloc_align()
4744 page = virt_to_page(nc->va); in page_frag_alloc_align()
4746 if (!page_ref_sub_and_test(page, nc->pagecnt_bias)) in page_frag_alloc_align()
4749 if (unlikely(nc->pfmemalloc)) { in page_frag_alloc_align()
4756 size = nc->size; in page_frag_alloc_align()
4762 nc->pagecnt_bias = PAGE_FRAG_CACHE_MAX_SIZE + 1; in page_frag_alloc_align()
4763 offset = size - fragsz; in page_frag_alloc_align()
4778 nc->pagecnt_bias--; in page_frag_alloc_align()
4780 nc->offset = offset; in page_frag_alloc_align()
4782 return nc->va + offset; in page_frag_alloc_align()
4808 while (page < --last) in make_alloc_exact()
4819 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4825 * allocate memory in power-of-two pages.
4847 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4873 * free_pages_exact - release memory allocated via alloc_pages_exact()
4892 * nr_free_zone_pages - count number of pages beyond high watermark
4899 * nr_free_zone_pages = managed_pages - high_pages
4917 sum += size - high; in nr_free_zone_pages()
4924 * nr_free_buffer_pages - count number of pages beyond high watermark
4940 zoneref->zone = zone; in zoneref_set_zone()
4941 zoneref->zone_idx = zone_idx(zone); in zoneref_set_zone()
4956 zone_type--; in build_zonerefs_node()
4957 zone = pgdat->node_zones + zone_type; in build_zonerefs_node()
4979 return -EINVAL; in __parse_numa_zonelist_order()
5000 * find_next_best_node - find the next node that should appear in a given node's fallback list
5038 /* Penalize nodes under us ("prefer the next node") */ in find_next_best_node()
5064 * This results in maximum locality--normal zone overflows into local
5065 * DMA zone, if any--but risks exhausting DMA zone.
5073 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; in build_zonelists_in_node_order()
5083 zonerefs->zone = NULL; in build_zonelists_in_node_order()
5084 zonerefs->zone_idx = 0; in build_zonelists_in_node_order()
5095 zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs; in build_thisnode_zonelists()
5098 zonerefs->zone = NULL; in build_thisnode_zonelists()
5099 zonerefs->zone_idx = 0; in build_thisnode_zonelists()
5116 /* NUMA-aware ordering of nodes */ in build_zonelists()
5117 local_node = pgdat->node_id; in build_zonelists()
5125 * distance group to make it round-robin. in build_zonelists()
5157 return zone_to_nid(z->zone); in local_memory_node()
5171 local_node = pgdat->node_id; in build_zonelists()
5173 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs; in build_zonelists()
5198 zonerefs->zone = NULL; in build_zonelists()
5199 zonerefs->zone_idx = 0; in build_zonelists()
5240 * trying to hold port->lock, for in __build_all_zonelists()
5242 * calling kmalloc(GFP_ATOMIC | __GFP_NOWARN) with port->lock held. in __build_all_zonelists()
5252 * building zonelists is fine - no need to touch other nodes. in __build_all_zonelists()
5254 if (self && !node_online(self->node_id)) { in __build_all_zonelists()
5269 * We now know the "local memory node" for each node-- in __build_all_zonelists()
5271 * Set up numa_mem percpu variable for on-line cpus. During in __build_all_zonelists()
5272 * boot, only the boot cpu should be on-line; we'll init the in __build_all_zonelists()
5273 * secondary cpus' numa_mem as they come on-line. During in __build_all_zonelists()
5274 * node/memory hotplug, we'll fixup all on-line cpus. in __build_all_zonelists()
5303 * (a chicken-egg dilemma). in build_all_zonelists_init()
5333 * more accurate, but expensive to check per-zone. This check is in build_all_zonelists()
5334 * made on memory-hotadd so a system can start with mobility in build_all_zonelists()
5368 * Clamp the batch to a 2^n - 1 value. Having a power in zone_batchsize()
5377 batch = rounddown_pow_of_two(batch + batch/2) - 1; in zone_batchsize()
5391 * can be a significant delay between the individual batches being in zone_batchsize()
5393 * fragmented and becoming unavailable for high-order allocations. in zone_batchsize()
5450 * pcp->high and pcp->batch values are related and generally batch is lower
5451 * than high. They are also related to pcp->count such that count is lower
5457 * store tearing. Any new users of pcp->batch, pcp->high_min and pcp->high_max
5459 * fully trust only the pcp->count field on the local CPU with interrupts
5469 WRITE_ONCE(pcp->batch, batch); in pageset_update()
5470 WRITE_ONCE(pcp->high_min, high_min); in pageset_update()
5471 WRITE_ONCE(pcp->high_max, high_max); in pageset_update()
5481 spin_lock_init(&pcp->lock); in per_cpu_pages_init()
5483 INIT_LIST_HEAD(&pcp->lists[pindex]); in per_cpu_pages_init()
5491 pcp->high_min = BOOT_PAGESET_HIGH; in per_cpu_pages_init()
5492 pcp->high_max = BOOT_PAGESET_HIGH; in per_cpu_pages_init()
5493 pcp->batch = BOOT_PAGESET_BATCH; in per_cpu_pages_init()
5494 pcp->free_count = 0; in per_cpu_pages_init()
5504 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); in __zone_set_pageset_high_and_batch()
5510 * Calculate and set new high and batch values for all per-cpu pagesets of a
5522 * PCP high is tuned manually, disable auto-tuning via in zone_set_pageset_high_and_batch()
5532 if (zone->pageset_high_min == new_high_min && in zone_set_pageset_high_and_batch()
5533 zone->pageset_high_max == new_high_max && in zone_set_pageset_high_and_batch()
5534 zone->pageset_batch == new_batch) in zone_set_pageset_high_and_batch()
5537 zone->pageset_high_min = new_high_min; in zone_set_pageset_high_and_batch()
5538 zone->pageset_high_max = new_high_max; in zone_set_pageset_high_and_batch()
5539 zone->pageset_batch = new_batch; in zone_set_pageset_high_and_batch()
5551 zone->per_cpu_zonestats = alloc_percpu(struct per_cpu_zonestat); in setup_zone_pageset()
5553 zone->per_cpu_pageset = alloc_percpu(struct per_cpu_pages); in setup_zone_pageset()
5558 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); in setup_zone_pageset()
5559 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); in setup_zone_pageset()
5584 pcp = per_cpu_ptr(zone->per_cpu_pageset, cpu); in zone_pcp_update_cacheinfo()
5587 * If data cache slice of CPU is large enough, "pcp->batch" in zone_pcp_update_cacheinfo()
5589 * consecutive high-order pages freeing without allocation. in zone_pcp_update_cacheinfo()
5591 * cache-hot pages sharing. in zone_pcp_update_cacheinfo()
5593 spin_lock(&pcp->lock); in zone_pcp_update_cacheinfo()
5594 if ((cci->per_cpu_data_slice_size >> PAGE_SHIFT) > 3 * pcp->batch) in zone_pcp_update_cacheinfo()
5595 pcp->flags |= PCPF_FREE_HIGH_BATCH; in zone_pcp_update_cacheinfo()
5597 pcp->flags &= ~PCPF_FREE_HIGH_BATCH; in zone_pcp_update_cacheinfo()
5598 spin_unlock(&pcp->lock); in zone_pcp_update_cacheinfo()
5632 memset(pzstats->vm_numa_event, 0, in setup_per_cpu_pageset()
5633 sizeof(pzstats->vm_numa_event)); in setup_per_cpu_pageset()
5638 pgdat->per_cpu_nodestats = in setup_per_cpu_pageset()
5649 zone->per_cpu_pageset = &boot_pageset; in zone_pcp_init()
5650 zone->per_cpu_zonestats = &boot_zonestats; in zone_pcp_init()
5651 zone->pageset_high_min = BOOT_PAGESET_HIGH; in zone_pcp_init()
5652 zone->pageset_high_max = BOOT_PAGESET_HIGH; in zone_pcp_init()
5653 zone->pageset_batch = BOOT_PAGESET_BATCH; in zone_pcp_init()
5656 pr_debug(" %s zone: %lu pages, LIFO batch:%u\n", zone->name, in zone_pcp_init()
5657 zone->present_pages, zone_batchsize(zone)); in zone_pcp_init()
5662 atomic_long_add(count, &page_zone(page)->managed_pages); in adjust_managed_page_count()
5691 * Perform a kasan-unchecked memset() since this memory in free_reserved_area()
5759 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
5770 pgdat->totalreserve_pages = 0; in calculate_totalreserve_pages()
5773 struct zone *zone = pgdat->node_zones + i; in calculate_totalreserve_pages()
5779 if (zone->lowmem_reserve[j] > max) in calculate_totalreserve_pages()
5780 max = zone->lowmem_reserve[j]; in calculate_totalreserve_pages()
5789 pgdat->totalreserve_pages += max; in calculate_totalreserve_pages()
5798 * setup_per_zone_lowmem_reserve - called whenever
5809 for (i = 0; i < MAX_NR_ZONES - 1; i++) { in setup_per_zone_lowmem_reserve()
5810 struct zone *zone = &pgdat->node_zones[i]; in setup_per_zone_lowmem_reserve()
5816 struct zone *upper_zone = &pgdat->node_zones[j]; in setup_per_zone_lowmem_reserve()
5821 zone->lowmem_reserve[j] = 0; in setup_per_zone_lowmem_reserve()
5823 zone->lowmem_reserve[j] = managed_pages / ratio; in setup_per_zone_lowmem_reserve()
5834 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10); in __setup_per_zone_wmarks()
5848 spin_lock_irqsave(&zone->lock, flags); in __setup_per_zone_wmarks()
5857 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN) in __setup_per_zone_wmarks()
5865 zone->_watermark[WMARK_MIN] = min_pages; in __setup_per_zone_wmarks()
5871 zone->_watermark[WMARK_MIN] = tmp; in __setup_per_zone_wmarks()
5883 zone->watermark_boost = 0; in __setup_per_zone_wmarks()
5884 zone->_watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp; in __setup_per_zone_wmarks()
5885 zone->_watermark[WMARK_HIGH] = low_wmark_pages(zone) + tmp; in __setup_per_zone_wmarks()
5886 zone->_watermark[WMARK_PROMO] = high_wmark_pages(zone) + tmp; in __setup_per_zone_wmarks()
5888 spin_unlock_irqrestore(&zone->lock, flags); in __setup_per_zone_wmarks()
5896 * setup_per_zone_wmarks - called when min_free_kbytes changes
5897 * or when memory is hot-{added|removed}
5978 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so in postcore_initcall()
6020 pgdat->min_unmapped_pages = 0; in setup_min_unmapped_ratio()
6023 zone->zone_pgdat->min_unmapped_pages += (zone_managed_pages(zone) * in setup_min_unmapped_ratio()
6048 pgdat->min_slab_pages = 0; in setup_min_slab_ratio()
6051 zone->zone_pgdat->min_slab_pages += (zone_managed_pages(zone) * in setup_min_slab_ratio()
6071 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6096 * percpu_pagelist_high_fraction - changes the pcp->high for each zone on each
6118 ret = -EINVAL; in percpu_pagelist_high_fraction_sysctl_handler()
6210 /* Usage: See admin-guide/dynamic-debug-howto.rst */
6234 .nid = zone_to_nid(cc->zone), in __alloc_contig_migrate_range()
6240 while (pfn < end || !list_empty(&cc->migratepages)) { in __alloc_contig_migrate_range()
6242 ret = -EINTR; in __alloc_contig_migrate_range()
6246 if (list_empty(&cc->migratepages)) { in __alloc_contig_migrate_range()
6247 cc->nr_migratepages = 0; in __alloc_contig_migrate_range()
6249 if (ret && ret != -EAGAIN) in __alloc_contig_migrate_range()
6251 pfn = cc->migrate_pfn; in __alloc_contig_migrate_range()
6254 ret = -EBUSY; in __alloc_contig_migrate_range()
6258 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone, in __alloc_contig_migrate_range()
6259 &cc->migratepages); in __alloc_contig_migrate_range()
6260 cc->nr_migratepages -= nr_reclaimed; in __alloc_contig_migrate_range()
6262 ret = migrate_pages(&cc->migratepages, alloc_migration_target, in __alloc_contig_migrate_range()
6263 NULL, (unsigned long)&mtc, cc->mode, MR_CONTIG_RANGE, NULL); in __alloc_contig_migrate_range()
6266 * On -ENOMEM, migrate_pages() bails out right away. It is pointless in __alloc_contig_migrate_range()
6269 if (ret == -ENOMEM) in __alloc_contig_migrate_range()
6275 if (!(cc->gfp_mask & __GFP_NOWARN) && ret == -EBUSY) in __alloc_contig_migrate_range()
6276 alloc_contig_dump_pages(&cc->migratepages); in __alloc_contig_migrate_range()
6277 putback_movable_pages(&cc->migratepages); in __alloc_contig_migrate_range()
6284 * alloc_contig_range() -- tries to allocate given range of pages
6286 * @end: one-past-the-last PFN to allocate
6313 .order = -1, in alloc_contig_range()
6338 * This lets us mark the pageblocks back as in alloc_contig_range()
6351 * In case of -EBUSY, we'd like to know which page causes problem. in alloc_contig_range()
6358 * -EBUSY is not accidentally used or returned to caller. in alloc_contig_range()
6361 if (ret && ret != -EBUSY) in alloc_contig_range()
6378 * We don't have to hold zone->lock here because the pages are in alloc_contig_range()
6407 ret = -EBUSY; in alloc_contig_range()
6414 ret = -EBUSY; in alloc_contig_range()
6420 free_contig_range(outer_start, start - outer_start); in alloc_contig_range()
6422 free_contig_range(end, outer_end - end); in alloc_contig_range()
6465 unsigned long last_pfn = start_pfn + nr_pages - 1; in zone_spans_last_pfn()
6471 * alloc_contig_pages() -- tries to find and allocate contiguous range of pages
6502 spin_lock_irqsave(&zone->lock, flags); in alloc_contig_pages()
6504 pfn = ALIGN(zone->zone_start_pfn, nr_pages); in alloc_contig_pages()
6514 spin_unlock_irqrestore(&zone->lock, flags); in alloc_contig_pages()
6519 spin_lock_irqsave(&zone->lock, flags); in alloc_contig_pages()
6523 spin_unlock_irqrestore(&zone->lock, flags); in alloc_contig_pages()
6533 for (; nr_pages--; pfn++) { in free_contig_range()
6560 __zone_set_pageset_high_and_batch(zone, zone->pageset_high_min, in zone_pcp_enable()
6561 zone->pageset_high_max, zone->pageset_batch); in zone_pcp_enable()
6570 if (zone->per_cpu_pageset != &boot_pageset) { in zone_pcp_reset()
6572 pzstats = per_cpu_ptr(zone->per_cpu_zonestats, cpu); in zone_pcp_reset()
6575 free_percpu(zone->per_cpu_pageset); in zone_pcp_reset()
6576 zone->per_cpu_pageset = &boot_pageset; in zone_pcp_reset()
6577 if (zone->per_cpu_zonestats != &boot_zonestats) { in zone_pcp_reset()
6578 free_percpu(zone->per_cpu_zonestats); in zone_pcp_reset()
6579 zone->per_cpu_zonestats = &boot_zonestats; in zone_pcp_reset()
6599 spin_lock_irqsave(&zone->lock, flags); in __offline_isolated_pages()
6627 spin_unlock_irqrestore(&zone->lock, flags); in __offline_isolated_pages()
6640 struct page *page_head = page - (pfn & ((1 << order) - 1)); in is_free_buddy_page()
6653 * Break down a higher-order page in sub-pages, and keep our target out of
6664 high--; in break_down_buddy_pages()
6693 spin_lock_irqsave(&zone->lock, flags); in take_page_off_buddy()
6695 struct page *page_head = page - (pfn & ((1 << order) - 1)); in take_page_off_buddy()
6708 __mod_zone_freepage_state(zone, -1, migratetype); in take_page_off_buddy()
6715 spin_unlock_irqrestore(&zone->lock, flags); in take_page_off_buddy()
6730 spin_lock_irqsave(&zone->lock, flags); in put_page_back_buddy()
6738 spin_unlock_irqrestore(&zone->lock, flags); in put_page_back_buddy()
6750 struct zone *zone = &pgdat->node_zones[ZONE_DMA]; in has_managed_dma()
6775 return -EINVAL; in accept_memory_parse()
6801 if (list_empty(&zone->unaccepted_pages)) in try_to_accept_memory_one()
6804 spin_lock_irqsave(&zone->lock, flags); in try_to_accept_memory_one()
6805 page = list_first_entry_or_null(&zone->unaccepted_pages, in try_to_accept_memory_one()
6808 spin_unlock_irqrestore(&zone->lock, flags); in try_to_accept_memory_one()
6812 list_del(&page->lru); in try_to_accept_memory_one()
6813 last = list_empty(&zone->unaccepted_pages); in try_to_accept_memory_one()
6815 __mod_zone_freepage_state(zone, -MAX_ORDER_NR_PAGES, MIGRATE_MOVABLE); in try_to_accept_memory_one()
6816 __mod_zone_page_state(zone, NR_UNACCEPTED, -MAX_ORDER_NR_PAGES); in try_to_accept_memory_one()
6817 spin_unlock_irqrestore(&zone->lock, flags); in try_to_accept_memory_one()
6835 to_accept = high_wmark_pages(zone) - in try_to_accept_memory()
6836 (zone_page_state(zone, NR_FREE_PAGES) - in try_to_accept_memory()
6844 to_accept -= MAX_ORDER_NR_PAGES; in try_to_accept_memory()
6864 spin_lock_irqsave(&zone->lock, flags); in __free_unaccepted()
6865 first = list_empty(&zone->unaccepted_pages); in __free_unaccepted()
6866 list_add_tail(&page->lru, &zone->unaccepted_pages); in __free_unaccepted()
6869 spin_unlock_irqrestore(&zone->lock, flags); in __free_unaccepted()